Google's experimental AI mode pushes search toward a closed ecosystem, mirroring OpenAI's ChatGPT search and Perplexity while potentially delivering a fatal blow to the open web.
The system, built on a modified version of Gemini 2.0 that has been specially adapted for this type of search, processes multiple related searches in parallel to handle complex, multi-part questions.
According to Google product manager Robby Stein, the technology combines Gemini 2.0's agentic capabilities with Google's information systems such as the Knowledge Graph, real-time data, and shopping information on billions of products to create comprehensible answers. Access requires a Google One AI Premium subscription, with initial testing limited to invited users through Google's Labs environment.
Video: Google
Alongside AI Mode, Google is upgrading its AI Overviews - which already serve over one billion users, according to Google - with its latest Gemini 2.0 models. These shorter AI responses will now handle programming, advanced mathematics, and multimodal queries better. The company is also expanding access to teenagers and removing registration requirements.
Preemptive damage control
Google, which has already faced ridicule for AI errors in its AI Overviews feature, says that the system defaults to traditional search results when uncertain about answers. Product manager Robby Stein is doing some upfront damage control, noting that "As with any early-stage AI product, we won't always get it right."
All AI answer engines have demonstrated such errors, but still, users might prefer getting direct answers over searching through sources themselves because it's convenient - even if they can't be sure if they are correct. Many probably won't even know that these AI systems can make mistakes. Questions also remain about accountability for AI-generated information, especially when errors affect individuals, as seen with Microsoft's Copilot search.
World Wide Wasteland
Stein tries to reassure site owners worried about traffic loss by claiming that "Helping people discover content from the web remains central to our approach, and with AI Mode we’re making it easy for people to explore and take action."
But reality belies this reassurance. Recent research from content-licensing company TollBit shows AI search engines like OpenAI and Perplexity reduce referral traffic to news sites and blogs by 96% compared to traditional Google search. Even a decline in the low double-digit percentage range could seriously impact the web ecosystem. Meanwhile, AI systems' website scraping has more than doubled in recent months.
Publishers dependent on website traffic face particular challenges from AI search. OpenAI has already begun splitting potential opponents using a "divide and conquer" strategy, luring major publishers with content licensing deals to power its answer engine embedded in ChatGPT. As AI increasingly mediates access to web content, this approach could further concentrate control over global media and opinion diversity in the hands of a few U.S. tech companies.
